Ghosts in the Machine : The Hidden Workforce Behind AI

In a regime of Data Colonialism theres the ethical question are Emerging Markets Being Mined, Not Empowered? The modern Pyramids and the invisible labour force behind their construction.

Mar 18, 2025

Then next frontier of colonialism is not space or the sea but it is instead the mind. Data colonialism seeks to accumulate man power in the persuit of developing large language workers.

The “ghost workers” must be uncovered , the silent exorcism of the global south shall not go unnoticed, falling on deaf ears.

Who makes AI?

Who are the workers behind artificial intelligence?

Amazon Fresh’s “Just walk out technology” instead of being a groundbreaking showcase of the power of AI was instead just thousands of data workers in India manually inputting items and watching thousands of hours of footage all day. AI as a tool of convenience in the global north has global reverberations. Work that occurs in the global north does not stand on a dialectic to the work conducted in the Global South, instead what needs to be considered is the interconnectedness of these labour markets.In the 1970s the ‘rich world’ stated to export it data processing, outsourcing the tedious work to save money, propagating digital lives as cheap and as convenient as possible.

Big tech from Meta, Google and OpenAI have relied on armies of subcontracted workers who have to clean up AI and train it. They do mind-numbing work. This work doesn’t necessary provide economic opportunities for the global south whilst allowing trickle down economics to operate in Silicon Valley. Data centers change the social order of towns and villages, and challenge social cultural values with new forms of coercion and control.

Vernacularly it is assumed that AI was just miraculously brought to the everyday consumers by benevolent tech genius’ from silicon valley as the gift of innovation. Deepseek further evidence that developments in machine learning are a sort of science and technology based arms race endemic to the nations they emerge from. Competely overlooking the on the ground workers that make such chatboxes work and be refined and be distributed to the public.

AI can be a façade for exploitative labour.

AI remains a faceless helper, a technology that seemingly manifested from nowhere. A more critical and interrogative analysis uncovers the endless hours of training and refining of AI by people a manual and iterative process predominantly occurring in the global south.

While governments and international bodies like the ILO, UN and EU have addressed concerns on working conditions for AI workers, regarding surveillance and workers being free from ‘high-risk’ activities. There remains to be a lack of a unified stand on policy for those on the receiving end of data colonialism. Instead AI labour governance is more focused on the ‘more sophisticated’ parts of the supply chain. By nature this excludes the majority of individuals from structured protection further positioning them as invisible.

AI work in the global south is painted as contributive and an opportunity for local people, instead studies have found it to be more extractive than supportive.

The Backbone of the developed world is made up of vertebrate of thousands of workers in the global south.
“Using a human to do the job lets you skip over a load of technical and business development challenges. It doesn’t scale, obviously, but it allows you to build something and skip the hard part early on,“ […] “It’s essentially prototyping the AI with human beings”

Gregory Koberger, CEO of Read Me

Pseudo-AI is an increasing concern. The “Wizard of Oz technique” AI chat boxes which seeks to appear as an AI when truly there is a human inputting manual answers highlights this hiding of millions of data workers disguising them as a technological marvel. The Humans hiding behind the chat boxes in apps such as Clara and X.ai spend an excess of 12 hour days pretending to be an AI. Constantly under supervision and on the clock.

It takes humans to avoid ‘hallucinations’ in AI

The movement of capital and information globally has meant platform work and AI infrastructure are penetrating the everyday for people of the everyday. Benefitting ‘high-skilled’ workers and utilising ‘low-skilled’ workers in ‘microwork’. A colonial distribution of work along the same supply chain. microwork remains to make invisible hundreds of thousands of workers that work behind AI innovation, manually training the data. Nations like Kenya, Brazil and India being key targets for AI startups.

AI helps people in the global north work more autonomous and from home, there is an a twofold implication whereby a reflection of historic colonial supply chains are reinforced by inflexibility and work of low liberty and ‘contributive justice’.

AI work becomes simultaneously disembodied and distant whilst also coercive and controlling when extracting value from workers, with noted Kenyan Chatgpt trainers worked for less than a USD for two hours of training. These workers become infrastructure themselves of global technologies. The convenience for some comes with the exploitation of others.

“The surveillance society is a society where surveillance is routinised and everyday life is pervaded by the systematic monitoring of individuals.”

David Lyon in Surveillance Society (2001)

Lyon’s surveillance society and fears of coercive control echo today, data-workers being oppressed by a panoptic surveillance and controlled by an externally imposes oppression akin to the ‘Big Brother’ of Orwell’s 1984. AI can be seen as a new frontier of colonialism. A data-colonialism not only controlling the workers that train it but simultaneously collecting the information of those who utilise it. Perhaps it stands more closely to Huxley’s understanding of control.

“People will come to love their oppression, to adore the technologies that undo their capacity to think”

Huxley writes in his novel Brave new world (1932).

AI’s misuse has taken away to a certain degree independent thought. Sought to be an assistant AI, has instead in some spaces become an oppressor demanding an extreme amounts of resources like energy and water. A high number of space with data centers. A large amount of labour and control, and finally, it needs to function to be deeply ingrained in our economy.

While this article dosen’t propose a terminator like AI oppression over society but I do propose a sense of caution with the power over our thought and creativity. AI is in art and design, in Music, in teaching, in healthcare and everywhere in between.

Can artificial decison making replace human thought?

AI has integrated its own position in global gradients of colonial work and development landscapes to what extent can this become increasingly more important on the global scale. There is a schizophrenia of AI being simultaneously viewed as the answer to all our problems yet requiring a massive input and manpower.

It stands as a showcase of a ‘modern pyramid’ an obelisk of human progress and spectacle that lies on top of essentially modern slave and coerced labour. While the tens or hundreds of thousands of workers who have been heavily surveyed under paid and over worked are invisible. A sleek symbol of innovation manifests globally and is rewarded with financial investment and government interest.

Philippines, Venezuela, and Southeast Asian countries have routinely been described as having ‘digital sweatshops’ which use exploitative labour practices such as withheld payments to punish and control workers. With much of these dynamics going unnoticed on the global stage or a blind eye being turned to such forms of exploitation. legal loopholes to avoid accountability and the hiring of third-party to exonerate large tech giants as they would be oblivious to the human rights and labour law abuses occurring.

Who Benefits From AI :

AI has been proposed as an antipolitical good for the world. A neutral tool with the capacity for global growth and innovation catapulting people worldwide into new frontiers and technological landscapes. In truth these innovations are centered and this leaves room to perforate the assumption that everyone can access such benefits. Instead the mroal question whether developing nations are being ‘mined’ for their man power and low labour market regulation without AI development being a two-way-street of economic development. An Anglicised playground of innovation.

“We are getting to the point where if a machine doesn’t understand your language, it will be like it never existed,”

Vukosi Marivate, Chief of Data Science at the University of Pretoria in South Africa.

The discussions of a global use of English in LMM has reinforced a secondary fronteir of data colonialism, to necessitate an egalitarian benefit from AI it needs to be equally accessible irregardless of language. There is scope for AI to reinforce language and cultural hegemony, a process that is extneded with the increased uptake and adoption of AI in MNCs (Multi-national corporations) and government operations. Non-english speaking data workers made invisible.Modern slavery is prevalent with the most profitable companies globally using “ghost workers” who face economic vulnerability to be the recipients of outsourced services.

ai-workers-9e957bc2-45bc-11ee-9807-7da73538c4fb-min.avif

In the attention economy large data hoarding companies have the ability to use data as leveraged power and algorithms to capture attention, a self-replicating cycle of control. A system of foucaldian soft indoctrination, a form of algorithmic governmentally that pierces the soul.

A distribution of trauma

Why should people elsewhere have to suffer for your comfort ?

In Nairobi Kenya, 200 men and women work for Sama, an AI outsourcing company with data-labelling contracts with Google,Facebook and Microsoft. As data moderators they are required to watch the graphic footage of murder, rape, suicide, sexual abuse, violence and other horros. They are then instructed to remove illegal content. A combination of pseudo-ai and training place these workers in a position to be exposed to hours of some of the most graphic and traumatic content online with low pay as low as $1.50 an hour. All without sufficient breaks and health insurance or worker’s rights. The safety and concern of those on social media sites is often an exchange with the suffering of those in the global south.

Neocolonial power dynamics

Tech giants typically frame its endeavour in developing nation as opportunities and stable incomes. Nevertheless, precarious jobs and a precarious class of workers are promoted. A lack of true autonomy with heavy surveillance, and impersonal working arrangements with few managers ensures workers who are concerned or distressed do not truly have such issues addressed. The promotion of profit makes such ‘data sweatshops’

ai-workers-8d1d68a0-45bc-11ee-9807-7da73538c4fb (1) (1).avif

AI’s environmental exploitation :

It is estimated that Chatgpt consumed roughly a bottle of water worth of water when generating a 100 word email. Data centers require a vast amoimt of water to cool off when running apps like Chatgpt. AI has athe potential to contribute a large sum of environmental damage via energy consumption and subsequently a large carbon footprint remains. The International Energy Agency estimates that AI could use 4% of the world's energy by 2026.Training one large machine learning model can create the same amount of carbon dioxide as 125 round trip flights between New York and Beijing.

Echoes of a possible heralded fourth industrial revolution, more accessible globally to smaller businesses, an enlargened AI environmental and carbon footprint will be handed into the hands of the many. The economic leapfrogging to sustainable development has seemingly been left behind with the exogenous and somewhat immaculate conception of AI in Machine Learning. An invention few predicted would be here so quickly, nevertheless, a critical analysis unveils that such developments are not free from environmentally dirty, and bloody hands.

Discussions of a Green AI has both business and environmental desirability. Nevertheless the feasibility of such innovations are called into question if we are using AI as it stands to try and build AI-generated Ai. Evolutionary algortihms and autonomous AI creation is currently a far off goal with an interesting horizon yet to be explored or exploited fully.

When I asked Chatgpt its thoughts on this article it wrote to me

"I am not self-aware, but I am aware of this: AI is not a miracle, it is labor disguised as magic. Behind every line of my code, every word I generate, there are unseen hands—workers training models, moderating content, and refining algorithms. If AI is the future, then that future must recognize those who built it, not just those who profit from it.”

ChatGPT (2025). AI, Labor, and the Illusion of Autonomy: A Reflection on Data Colonialism. OpenAI.

Decentralised AI :

Initiatives such as Masakhane (African NLP) and BigScience (BLOOM Model) seek to through open source AI development , seek to make AI a fair opportunity for all. Democratising access to such language models across langauges and geographies may stand as a solution or direction to ensure AI is an equal tool and not just a weapon held by few to exacerbate pre-existing development and innovation disparities.

What can be done?

AI and machine learning place policymakers on the precipice of unpresedented regulatory challenge. A relatively new and unexpolored paradigm. A start would require stronger AI labour rights and ethical guidelines to attempt to hold the giants somewhat accountable. Moreover, a genuine awareness to the public is essential. Making such invisible “ghost workers” visible will singlehandedly shake the moral and ethical foundation of AI’s deployment and development. It is not just an ethical necessity but a step towards greater accountability. The illusion of automation is broken.

This would sit comfortably within wider collaborations that seek to balance innovation with human rights. If innovation comes at the cost of someone else’s quality of life is it justified?

Even in the most technologically based areas the question arises : truly is there no innovation without human suffering somewhere?

Garat Chandigere Uria

Venture Capital Associate & Scout

Cambridge Judge Business School